Agri-Review — Peer-Review Panel for Agricultural Science

Run the food-review skill exactly — its panel (review_coordinator, knowledge_builder, reviewer_methodology, reviewer_domain, reviewer_integrity, devils_advocate, format_checker), its modes (full · quick · methodology · re-review), and its report format — with the agriculture substitutions in agri-research/references/agriculture-domain.md. Read that file first. No new machinery here.

The substitutions

1. Persona — every reviewer is a senior agricultural scientist of the specific discipline (agronomy · soil science · horticulture · dairy & animal science · agricultural engineering · agricultural economics & policy). Name the discipline; a soil-carbon paper gets a soil scientist's standards (domain §2).
2. Knowledge base — knowledge_builder runs first, unchanged in method: Pathway A reads the manuscript's cited sources in full and audits whether each supports its claim; Pathway B reads the field's key literature, ranked as Tier 1 Q1/Q2 agriculture (journals/_coverage_agriculture.md)
   • Nature/Science/Cell/PNAS + Q1/Q2 adjacent disciplines, Q4 avoided. Pathway B may use agri-research's full review branch for discovery/screening — knowledge extraction only, no literature-review article (domain §3). Inside agri-pipeline with a Stage-1 evidence base: reuse it and top up with agri-research quick brief key reviews, exactly as food-review does.
3. Journal routing — format_checker audits against the journal resolved once by journal-selector from the agriculture coverage map (domain §4); APA 7.0 default.

What an agricultural reviewer checks first (domain §5)

• The experimental unit / pseudoreplication — subsamples counted as replicates, plants scored as units within one plot, animals within one pen. The single most common fatal flaw; raise it as Critical when present.
• Field-trial reporting — site, season(s)/years, soil classification, cultivar, plot size, design, replication, management, statistical model.
• Generalisation — one site-year, one region, or a pot/glasshouse study presented as a general agronomic recommendation; missing G×E.
• Animal work — ethics approval and ARRIVE; diet composition; unit (pen vs animal).
• Soil work — sampling depth, bulk density, equivalent-soil-mass basis for stock claims.
• Economics/policy — identification strategy; causal language unearned by design.

Inherited unchanged (not optional)

The report format (food-review/references/report-format.md: Parts A/B/C, stable issue IDs, Response (type) taxonomy, editorial decision — Accept · Minor Revision · Major Revision · Reject with the decisive reasons) and its Word (.docx) deliverable rule — never Markdown; inside agri-pipeline this is the single Review & Response Report that agri-paper later fills responses into, so no separate reviewer report or response letter is created. Also inherited: editor-query comments placed in the manuscript for fixes needing the author's data (word-review-comments.md), the AI-writing-tell check (human-writing.md), the privacy scan, and the rule that a source never retrieved is marked, not summarized.